Sie sind auf Seite 1von 10

Using analytics

for insurance
fraud detection
3 innovative methods and a
10-step approach to kick
start your initiative

Ruchi Verma
Sathyan Ramakrishna Mani

If youve been used to thinking about analytics in terms of


sales or marketing, think again. Today, analytics can reinvent
your enterprise technologies social networking, big data,
CRM to crack down on financial offenders. Giving you
more than an insight a day, to keep the fraud away.

Digitization

a new
opportunity for
fraud detection?
D
igitization marked by a growing number of mobile
devices and social media is changing the business
landscape for all sectors including insurance. The
opportunities offered by this landscape for insurers are vast.
Social networks and communities help insurers connect with
their customers better, which in turn aids branding, customer
acquisition, and retention. Insurance firms also receive a
plethora of inputs from digital information in the form of
feedback, which also can be used to come up with customized
products and competitive pricing.

The big data trend, (the


growth in unstructured
data) always leaves lots
of room for a fraud going
undetected if data is not
analyzed thoroughly

In addition to these opportunities, insurance companies


are harnessing digitization using data analytics for fraud
detection. Handling fraud manually has always been costly
for insurance companies, even if one or two low incidences
of high-value fraud went undetected. In addition to this,
the big data trend, (the growth in unstructured data) always
leaves lot of room for a fraud going undetected if data is not
analyzed thoroughly.

Using analytics for insurance fraud detection

Digital Transformation

Traditionally, insurance
companies use statistical
models to identify
fraudulent claims

Fraud
detection by
insurance
companies

These models have their own disadvantages. First, they use sampling methods to analyze
data, which leads to one or more frauds going undetected. There is a penalty for not
analyzing all the data. Second, this method relies on the previously existing fraud cases, so
every time a new fraud occurs, insurance companies have to bear the consequences of the
first time. Finally, the traditional method works in silos and is not quite capable of handling
the ever-growing sources of information from different channels and different functions in an
integrated way.
Analytics addresses these challenges and plays a very crucial role in fraud detection for
insurance companies. Some of the key benefits of using analytics in fraud detection are
discussed below.
Using sampling techniques comes with its own set of accepted errors. By using analytics, insurance
companies can build systems that run through all critical data. This in turn helps detect lowincidence (0.001%) events. Techniques such as predictive modeling can be used to thoroughly
analyze instances of fraud, filter obvious cases, and refer low-incidence fraud cases for further
analysis.

Analytics help in building a truly global perspective of the anti-fraud efforts throughout the
enterprise. Such a perspective often leads to effective fraud detection by linking associated
information within the organization. Fraud can occur at a number of source points: claims or
surrender, premium, application, employee-related or third-party fraud. At the same time, insurance
channel diversification is adding to the fragmentation of traceable data. Insurance-related activities
can be done via mobile devices apart from the traditional online and face-to-face insurance. This
can be viewed as an addition to information silos in the insurance industry. Given greater channel
diversification and the increase in areas where fraud can occur, it is important for insurers to have
accessible enterprise-level information about their business and customers.

Analytics plays an important role in integrating data. Effective fraud detection capabilities can
be built by combining data from various sources. Analytics also help in integrating internal data
with third-party data that may have predictive value, such as public records. Data sources with
derogatory attributes are all public records that can be integrated into a model. Examples include
bankruptcies, liens, judgements, criminal records, foreclosures, or even address change velocity
to indicate transient behavior. Other types of third-party data can be beneficial in enhancing
efficiencies such as review of appraisal information to determine if damages match description or
loss or injuries being claimed. One of the most under-utilized data sources is medical bill review
data. This data, if used in a model properly, is a gold mine for companies investigating medical
fraud. Uncovering anomalies, in billing and adding these to the other scoring engines or social
network analysis will decrease the amount of time an investigator or analyst spends trying to pull all
of the pieces together to identify fraudulent activity.

Using analytics for insurance fraud detection

Digital Transformation

Analytics helps in deriving the best value from unstructured data. Fraud can be soft fraud or hard
fraud. This is based on whether it consists of a policyholders exaggerated claims, or if it consists
of a policy holder planning or inventing a loss. At a high level, fraud can occur during commission
rebating, due to false documentation, collusion between parties or from mis-selling. Although lots
of structured information is stored in a data warehouse as part of many applications, most of the
crucial information about a fraud is in unstructured data, such as third party reports, which are
hardly analyzed. In most insurance firms, information available in social media is not appropriately
stored. A special-investigative-unit investigator will agree that unstructured data is very important
for fraud analysis. Since textual data is not directly used for reporting, it does not find a place in
most data warehouses. This is where text analytics can play a key role in reviewing this unstructured
data and providing some valuable insights in fraud detection.

Three innovative fraud detection methods


1. Social Network Analysis (SNA)
Lets take an example to explain the use of social network analysis (SNA). In a car accident, all people in the vehicle have
exchanged addresses and phone numbers and provided them to the insurer. However, the address given by one of the
accident victims may have many claims or the driven vehicle may have been involved in other claims. Having the ability
to cull this information saves time and gives the insurer an insight into the parameters involved in the fraud case. SNA
allows the company to proactively look through large amounts of data to show relationships via links and nodes.
The SNA tool combines a hybrid approach of analytical methods. The hybrid approach includes organizational business
rules, statistical methods, pattern analysis, and network linkage analysis to really uncover the large amounts of data to
show relationships via links. When one looks for fraud in a link analysis, one looks for clusters and how those clusters
link to other clusters. Public records such as judgments, foreclosures, criminal records, address change frequency, and
bankruptcies are all data sources that can be integrated into a model.
Using the hybrid approach, the insurer can rate these claims. If the rating is high, it indicates that the claim is fraudulent.
This may be because of a known bad address or suspicious provider or vehicle in many accidents with multiple carriers.

SNA follows this path:

Operational
data store

Extract
transform
load

Fraud
repostitory

1. The data (structured and unstructured) from


various sources is fed into the extract transform
and load tool. It is then transformed and loaded
into a data warehouse.
2. The analytics team uses information across a wide
variety of sources and scores the risk of fraud and
prioritizes the likelihood based on multiple factors.
The information used can range anywhere from a
prior conviction, a relationship in some manner
to another individual with a prior case, multiple
rejected claims, odd combinations of data, or even
odd modifications to personal information.
3. Technologies such as text mining, sentiment
analysis, content categorization and social network
analysis are integrated into the fraud identification
and predictive modeling process.
4. Depending on the score of the particular network,
an alert is generated.
5. The investigators can then leverage this
information and begin researching more on the
fraudulent claim.
6. Finally, issues or frauds that are identified are
added into the business use case system, which is
a part of the hybrid framework.

Using analytics for insurance fraud detection

Insurance fraud detection using social network analysis

Before implementing SNA, insurers should consider:


1. How fast data arrives
2. How clean the data is when it arrives
3. How deep the analysis must go to get the results
4. What type of user interface components need to be included
in the SNA dashboard

Digital Transformation

Case study: GE Consumer & Industrial Home Services Division


Scenario
In GE Consumer & Industrial Home Services Division, claims typically came from technicians who repair consumer products
that are under warranty. One of the biggest problems with their old process was that they could not identify patterns. With the
amount of data available to them, no one could see unusual behavior emerging. Sometime back, GE got the perfect scenario
to test an SNA solution from SAS, a developer of business analytics software. The company was tipped off to some service
providers committing fraud. This situation made for an ideal pilot scenario. SAS was given the responsibility of analyzing the
available data and identifying patterns in the data to find out who was committing the fraud.
Functioning of the fraud detection system
Typically, there are some metrics and indicators on every claim that assist in identifying suspicious or fraudulent claims. GEs
claims data is fed into the fraud detection software. There are 26 claim-level analyses, which are automatically calculated for
each claim. There are some indicators like flags that are calculated based on various metrics and sent for auditing when they
indicate that multiple elements in the claim fall out of the normal curve. Once these claims are flagged, the auditors at GE
investigate these suspicious claims.
Outcome
The GE Consumer & Industrial Home Services Division estimated that it saved about $5.1 million in the first year of using SAS,
to detect suspect claims.

2. Predictive analytics for big data


Consider a scenario when a person raises a claim saying that his car caught fire, but the story that was narrated by him indicates
that he took most of the valuable items out prior to the incident. That might indicate the car was torched on purpose. Predictive
analytics include the use of text analytics and sentiment analysis to look at big data for fraud detection. Claim reports span across
multiple pages, leaving very little room for text analytics to detect the scam easily. Big data analytics helps in sifting through
unstructured data, which wasnt possible earlier and helps in proactively detecting frauds. There has been an increase in the use of
predictive analytics technology, which is a part of big data analytics concept, to spot potentially fraudulent claims and speed the
payment of legitimate ones. In the past, predictive analytics were used to analyze statistical information stored in the structured
databases, but now it is branching out into the big data realm. The potential fraud present in the written report above is spotted
using text analytics and sentiment analysis.
Heres how the text analytics technology
works:
Claim adjusters write long reports when
they investigate the claims
Clues are normally hidden in the reports,
which the claims adjuster would not have
noticed
However, the computing system, which is
based on business rules, can spot evidence
of possible fraud
The most important point to observe is
that people who usually commit fraud alter
their story over time. The fraud detection
system can spot these discrepancies

Using analytics for insurance fraud detection

Digital Transformation

Case study: Infinity Insurance Co.


Infinity, a property and casualty company, came up with the idea of scoring insurance claims from customers to look for signs
of fraud. Its target market is mainly drivers who have higher than normal risks and pay high rates compared to others. With
the kind of exposure Infinity has, spotting insurance fraud, either while raising the claim or while calculating the premium to
be paid, is even more important than it is to other insurance companies. Infinity uses a predictive analytics technology to spot
potentially fraudulent claims and speed the payment of legitimate ones.
After using predictive analysis, the claims fraud system increased the success rate in pursuing fraudulent claims from 5088 %
and reduced the time required to refer questionable claims for investigation by as much at 95%.

3. Social customer relationship management (CRM)


Social CRM is neither a platform nor a technology, but rather, a process. It is important
that insurance companies link social media to their CRM. When social media is
integrated within multiple layers of the organization, it enables greater transparency
with customers. Mutually beneficial transparency indicates that the company trusts
its customers and vice versa. This customer-centric ecosystem reinforces the fact
that increasingly the customer is in control. This customer-centric ecosystem can
be beneficial to the business as well, if the business is able to leverage the collective
intelligence of its customer base.
Social CRM uses a companys existing CRM and gathers data from various social
media platforms. It uses a listening tool to extract data from social chatter, which
acts as reference data for the existing data in the current CRM. The reference data
along with information stored in the CRM is fed into a case management system. The
case management system then analyzes the information based on the organizations
business rules and sends a response. The response from the claim management
system as to whether the claim is fraudulent or not, is then confirmed by investigators
independently, since the output of social analytics is just an indicator and should not
be taken as the final reason to reject a claim.

Regulators

Customer
Business

Case study: AXA OYAK, Turkey


AXA OYAK is a Turkish insurance company that has been using the SAS Social CRM solution to manage risk and prevent fraud. AXA
OYAK built an intelligent enterprise around social CRM in such a way that it integrates all customer-related information into a single
and coordinated corporate vision.
Using its social CRM, AXA was able to clean up their customer portfolio data. This helped them find and correct inconsistencies
in this data, which enables AXA to link two slightly different records to the same customer. With cleaner data, AXA can run more
accurate customer analysis and investigate fraudulent claims more efficiently. Using SAS, AXA OYAK was quickly able to find
the relationships between customer behavior and fraudulent claims. With the SAS data warehouse, AXA is able to segment their
customer data based on flags that are generated while analyzing certain relationships between data sets.

Using analytics for insurance fraud detection

Digital Transformation

A 10-step
approach to
implement
analytics for
fraud detection

Perform SWOT

2 Build a dedicated
fraud management
team

Whether to build
or buy

Clean data

Come up with
relevant business
rules

Many insurance fraud detection tools target only a specific insurance vertical, such as claim
management, and build the entire framework around it. For making the insurance fraud
framework more robust, a more holistic framework is needed. One which examines all potential
areas for fraud claims, premiums, applications, employee and vendor details in an integrated
fashion. Here we outline 10 steps for implementing analytics for fraud detection.

Insurance companies are realizing the importance of analytics in the fraud detection space
and hurriedly opting for expensive fraud solutions that are not aligned to the companys
weakness and strengths. In order to leverage analytics solutions to the fullest, insurance
companies should first do a SWOT analysis of existing fraud detection frameworks and
processes to identify gaps.

Usually, in a traditional insurance company, no specific team or person is proactively


accountable for fraud detection. When fraud is detected internally, people point fingers, raise
alarms and take measures to fight it. It is important that a dedicated team is identified and
made accountable for fraud detection. The team should report to senior management for
necessary buy in.

Once the SWOT is complete and a team of dedicated people for fraud detection have been
identified, insurance companies should review how they want to implement analytics and
what data sources they want to analyze. Insurance firms need to be honest in answering
whether the skill set for building analytics solutions are available in-house or whether there
is a need to buy an analytical fraud detection solution from an external vendor. If there is
a need to buy the analytics solution, insurance firms should evaluate different analytics
vendors in the market to find a solution that best fits the companys requirements. Key
parameters to judge an external vendor are cost, user interface, scalability, ease of integration
and ability to add new data sources.

Integrate siloed databases and remove inefficiencies from processes and redundancies from
data sources.

Insurance companies should leverage existing domain expertise and experienced resources to
come up with business rules. Certain types of fraud are very specific to the industry and, in
some cases, certain companies. Without inputs from in-house capabilities, it will be difficult
for any internal or external team to build a robust fraud detection solution.

Using analytics for insurance fraud detection

Digital Transformation

6 Come up with

pre-determined
anomaly detection
thresholds

7 Use predictive
modeling

Whether the analytics framework is built in-house or by using a third-party vendor, insurance
companies should provide inputs for threshold values for different anomalies. The number
of claims received for life insurance is different from the number of claims received in nonlife insurance. Key performance indicators associated with tasks or events are baselined and
thresholds are set using anomaly detection. Setting the threshold is a major decision in anomaly
detection. If thresholds are set too high, too many fraudulent claims could slip through the
system. When thresholds are set too low, there can be risks of wasting time, alienating members
and providers, and can result in late-payment penalties. Certain statistical analyses take an
empirical value by determining normal ranges for predetermined metrics.

An important fraud detection method is one that utilizes data mining tools to build models that
produce fraud propensity scores linked to unidentified metrics. Claims are automatically scored
to look for any indication of a discrepancy or fraud. After this, the results are made available for
review and further analysis.

8 Use of SNA

SNA has proven effective in identifying organized fraud activities by modeling relationships
between various entities involved in the claim. Entities can range anywhere between locations to
telephone numbers. The number of linkages between certain types of entities may be found to be
much greater than the average number of connections expected based on statistical analyses of
other networks of entities.

9 Build an integrated

Integrated case management capabilities allow investigators to capture all key findings that are
relevant to an investigation, including claims data, network diagrams, adjuster notes, and social
media, which can contain structured or unstructured data. Metrics are the key indicators of
fraud or abuse and can be automatically tabulated for comparison at the individual entity or
network level (using the anomaly threshold or SNA). Case workflow enables a full and complete
assessment of investigative workload, efficiency, and return on investment.

10

Insurance companies should keep looking for additional sources of data and integrate those with
existing fraud detection solutions, for building the most efficient fraud detection system possible
to address a variety of new frauds that may emerge in the future.

case management
system leveraging
social media

Forward-looking
analytics solutions

Using analytics for insurance fraud detection

Digital Transformation

The proposed system can


Rapidly organize and analyze the unstructured data present in the claims submitted by the claimant, notes of the claim adjuster
and third-party reports
Examine the sentiments of the claimant to help drill down to the specific concerns that bother at-risk customers
Synthesize complex fraudulent patterns that contain the presence of multiple red flag indicators
Detect and provide early warning of potential issues before they become problems
Uncover early patterns in fraudulent activity

The way
forward

Insurance firms always hesitate in implementing analytics because of the initial time investment
needed for analytics solutions. However, it has been seen that analytics goes a long way in
detecting fraud proactively and earlier in the insurance lifecycle. It culminates in reducing the
overall cost of fraud detection and improving the overall ROI of insurance fraud solutions.
Insurers must now exploit the existing data in any form (structured or unstructured) by using
analytics to effectively detect, manage, and report frauds. The earlier the fraud is detected in
the insurance lifecycle, the lesser it costs to manage it. Analytics can play a very important role
in identifying fraud early in the insurance lifecycle, and failing to act on this opportunity could
quite literally equate to a gargantuan loss.

About the
Authors

She has around eight and half years of experience in Infosys


in varied roles across multiple accounts. Her areas of interest
includes emerging trends and regulations in the financial
services and insurance domain.
She can be reached at ruchi_verma@infosys.com

Ruchi Verma
Senior Consultant, Financial
Services and Insurance Unit

He has close to three years of experience in varied roles across


multiple accounts. His interests are in the area of capital
markets. He is also a keen follower of macroeconomic events
that take place around the world.
He can be reach at sathyan_mani@infosys.com

Sathyan Ramakrishna Mani


Senior Associate Consultant,
Financial Services and
Insurance Unit
Using analytics for insurance fraud detection

Digital Transformation

Other articles in this edition of FINsights


Like this
article?
Share it
with your
network

Serve them all,


serve them well

Multi-channel user
experience a banking
perspective

Banking on customer
satisfaction in a
digital world

Enhancing customer
experience with
immersive
correspondence

1VUUJOHBOBNFUP
BGBDF


.POFUJ[BUJPOoMPPLJOH
BUUIFGVUVSFPG
TFMGTFSWJDFCBOLJOH

More the merrier:


the hackers motto

Financial transactions
get personalized and
secure with biometrics

Go digital, reduce fraud

Financial institutions
reduce fraud risk with
social media

Reinventing bank
marketing with mobility

Delivering on the go
services for employees
and customers

Just what the insurer


wanted: a tabletized
future

The social makeover of


the financial services
industry

Digital transformation
framework

Straight-through
processing

Read previous
article

10

Using analytics for insUrance fraUD Detection

Read next
article
Digital transformation